def add_task_arguments(cls, parser: argparse.ArgumentParser):
group = parser.add_argument_group(description="Task related")
# NOTE(kamo): add_arguments(..., required=True) can't be used
# to provide --print_config mode. Instead of it, do as
# required = parser.get_default("required")
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetEnhancementModel),
help="The keyword arguments for model class.",
)
group.add_argument(
"--criterions",
action=NestedDictAction,
default=[
{
"name": "si_snr",
"conf": {},
"wrapper": "fixed_order",
"wrapper_conf": {},
},
],
help="The criterions binded with the loss wrappers.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=False,
help="Apply preprocessing to data or not",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def add_task_arguments(cls, parser: argparse.ArgumentParser):
group = parser.add_argument_group(description="Task related")
group.add_argument(
"--num_spk",
type=int_or_none,
default=None,
help=
"The number fo speakers (for each recording) used in system training",
)
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--input_size",
type=int_or_none,
default=None,
help="The number of input dimension of the feature",
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetDiarizationModel),
help="The keyword arguments for model class.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=True,
help="Apply preprocessing to data or not",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def add_arguments(self, parser):
parser.add_argument(
f"--{self.name}",
type=lambda x: str_or_none(x.lower()),
default=self.default,
choices=self.choices(),
help=f"The {self.name} type",
)
parser.add_argument(
f"--{self.name}_conf",
action=NestedDictAction,
default=dict(),
help=f"The keyword arguments for {self.name}",
)
def add_task_arguments(cls, parser: argparse.ArgumentParser):
group = parser.add_argument_group(description="Task related")
# NOTE(kamo): add_arguments(..., required=True) can't be used
# to provide --print_config mode. Instead of it, do as
required = parser.get_default("required")
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--input_size",
type=int_or_none,
default=None,
help="The number of input dimension of the feature",
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(HynetImgrModel),
help="The keyword arguments for model class.",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def add_task_arguments(cls, parser: argparse.ArgumentParser):
# NOTE(kamo): Use '_' instead of '-' to avoid confusion
assert check_argument_types()
group = parser.add_argument_group(description="Task related")
# NOTE(kamo): add_arguments(..., required=True) can't be used
# to provide --print_config mode. Instead of it, do as
required = parser.get_default("required")
required += ["token_list"]
group.add_argument(
"--token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token",
)
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetLanguageModel),
help="The keyword arguments for model class.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=False,
help="Apply preprocessing to data or not",
)
group.add_argument(
"--token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word"],
help="",
)
group.add_argument(
"--bpemodel",
type=str_or_none,
default=None,
help="The model file fo sentencepiece",
)
parser.add_argument(
"--non_linguistic_symbols",
type=str_or_none,
help="non_linguistic_symbols file path",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
assert check_return_type(parser)
return parser
def add_task_arguments(cls, parser: argparse.ArgumentParser):
group = parser.add_argument_group(description="Task related")
# NOTE(kamo): add_arguments(..., required=True) can't be used
# to provide --print_config mode. Instead of it, do as
required = parser.get_default("required")
required += ["token_list"]
group.add_argument(
"--token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token",
)
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--input_size",
type=int_or_none,
default=None,
help="The number of input dimension of the feature",
)
group.add_argument(
"--ctc_conf",
action=NestedDictAction,
default=get_default_kwargs(CTC),
help="The keyword arguments for CTC class.",
)
group.add_argument(
"--asr_model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetASRModel),
help="The keyword arguments for model class.",
)
group.add_argument(
"--enh_model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetEnhancementModel),
help="The keyword arguments for model class.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=False,
help="Apply preprocessing to data or not",
)
group.add_argument(
"--token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word", "phn"],
help="The text will be tokenized " "in the specified level token",
)
group.add_argument(
"--bpemodel",
type=str_or_none,
default=None,
help="The model file of sentencepiece",
)
parser.add_argument(
"--non_linguistic_symbols",
type=str_or_none,
help="non_linguistic_symbols file path",
)
parser.add_argument(
"--cleaner",
type=str_or_none,
choices=[None, "tacotron", "jaconv", "vietnamese"],
default=None,
help="Apply text cleaning",
)
parser.add_argument(
"--g2p",
type=str_or_none,
choices=[None, "g2p_en", "pyopenjtalk", "pyopenjtalk_kana"],
default=None,
help="Specify g2p method if --token_type=phn",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def add_task_arguments(cls, parser: argparse.ArgumentParser):
group = parser.add_argument_group(description="Task related")
# NOTE(kamo): add_arguments(..., required=True) can't be used
# to provide --print_config mode. Instead of it, do as
required = parser.get_default("required")
required += ["src_token_list", "token_list"]
group.add_argument(
"--token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token (for target language)",
)
group.add_argument(
"--src_token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token (for source language)",
)
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--input_size",
type=int_or_none,
default=None,
help="The number of input dimension of the feature",
)
group.add_argument(
"--ctc_conf",
action=NestedDictAction,
default=get_default_kwargs(CTC),
help="The keyword arguments for CTC class.",
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetSTModel),
help="The keyword arguments for model class.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=True,
help="Apply preprocessing to data or not",
)
group.add_argument(
"--token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word", "phn"],
help="The target text will be tokenized "
"in the specified level token",
)
group.add_argument(
"--src_token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word", "phn"],
help="The source text will be tokenized "
"in the specified level token",
)
group.add_argument(
"--bpemodel",
type=str_or_none,
default=None,
help="The model file of sentencepiece (for target language)",
)
group.add_argument(
"--src_bpemodel",
type=str_or_none,
default=None,
help="The model file of sentencepiece (for source language)",
)
parser.add_argument(
"--non_linguistic_symbols",
type=str_or_none,
help="non_linguistic_symbols file path",
)
parser.add_argument(
"--cleaner",
type=str_or_none,
choices=[None, "tacotron", "jaconv", "vietnamese"],
default=None,
help="Apply text cleaning",
)
parser.add_argument(
"--g2p",
type=str_or_none,
choices=g2p_choices,
default=None,
help="Specify g2p method if --token_type=phn",
)
parser.add_argument(
"--speech_volume_normalize",
type=float_or_none,
default=None,
help="Scale the maximum amplitude to the given value.",
)
parser.add_argument(
"--rir_scp",
type=str_or_none,
default=None,
help="The file path of rir scp file.",
)
parser.add_argument(
"--rir_apply_prob",
type=float,
default=1.0,
help="THe probability for applying RIR convolution.",
)
parser.add_argument(
"--noise_scp",
type=str_or_none,
default=None,
help="The file path of noise scp file.",
)
parser.add_argument(
"--noise_apply_prob",
type=float,
default=1.0,
help="The probability applying Noise adding.",
)
parser.add_argument(
"--noise_db_range",
type=str,
default="13_15",
help="The range of noise decibel level.",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def test_str_or_none(value: str, desired: Any):
with pytest_raise_or_nothing(desired):
assert str_or_none(value) == desired
def add_task_arguments(cls, parser: argparse.ArgumentParser):
"""Add Transducer task arguments.
Args:
cls: ASRTransducerTask object.
parser: Transducer arguments parser.
"""
group = parser.add_argument_group(description="Task related")
required = parser.get_default("required")
required += ["token_list"]
group.add_argument(
"--token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token",
)
group.add_argument(
"--input_size",
type=int_or_none,
default=None,
help="The number of input dimension of the feature",
)
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"chainer_espnet1",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetASRTransducerModel),
help="The keyword arguments for model class.",
)
group.add_argument(
"--encoder_conf",
action=NestedDictAction,
default={},
help="The keyword arguments for encoder class.",
)
group.add_argument(
"--joint_network_conf",
action=NestedDictAction,
default={},
help="The keyword arguments for joint network class.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=True,
help="Apply preprocessing to data or not",
)
group.add_argument(
"--token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word", "phn"],
help="The text will be tokenized "
"in the specified level token",
)
group.add_argument(
"--bpemodel",
type=str_or_none,
default=None,
help="The model file of sentencepiece",
)
parser.add_argument(
"--non_linguistic_symbols",
type=str_or_none,
help="non_linguistic_symbols file path",
)
parser.add_argument(
"--cleaner",
type=str_or_none,
choices=[None, "tacotron", "jaconv", "vietnamese"],
default=None,
help="Apply text cleaning",
)
parser.add_argument(
"--g2p",
type=str_or_none,
choices=g2p_choices,
default=None,
help="Specify g2p method if --token_type=phn",
)
parser.add_argument(
"--speech_volume_normalize",
type=float_or_none,
default=None,
help="Scale the maximum amplitude to the given value.",
)
parser.add_argument(
"--rir_scp",
type=str_or_none,
default=None,
help="The file path of rir scp file.",
)
parser.add_argument(
"--rir_apply_prob",
type=float,
default=1.0,
help="The probability for applying RIR convolution.",
)
parser.add_argument(
"--noise_scp",
type=str_or_none,
default=None,
help="The file path of noise scp file.",
)
parser.add_argument(
"--noise_apply_prob",
type=float,
default=1.0,
help="The probability applying Noise adding.",
)
parser.add_argument(
"--noise_db_range",
type=str,
default="13_15",
help="The range of noise decibel level.",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --decoder and --decoder_conf
class_choices.add_arguments(group)
def add_task_arguments(cls, parser: argparse.ArgumentParser):
# NOTE(kamo): Use '_' instead of '-' to avoid confusion
assert check_argument_types()
group = parser.add_argument_group(description="Task related")
# NOTE(kamo): add_arguments(..., required=True) can't be used
# to provide --print_config mode. Instead of it, do as
required = parser.get_default("required")
required += ["token_list"]
group.add_argument(
"--token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token",
)
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetLanguageModel),
help="The keyword arguments for model class.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=True,
help="Apply preprocessing to data or not",
)
group.add_argument(
"--token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word"],
help="",
)
group.add_argument(
"--bpemodel",
type=str_or_none,
default=None,
help="The model file fo sentencepiece",
)
parser.add_argument(
"--non_linguistic_symbols",
type=str_or_none,
help="non_linguistic_symbols file path",
)
parser.add_argument(
"--cleaner",
type=str_or_none,
choices=[None, "tacotron", "jaconv", "vietnamese"],
default=None,
help="Apply text cleaning",
)
parser.add_argument(
"--g2p",
type=str_or_none,
choices=[
None,
"g2p_en",
"g2p_en_no_space",
"pyopenjtalk",
"pyopenjtalk_kana",
"pyopenjtalk_accent",
"pyopenjtalk_accent_with_pause",
"pypinyin_g2p",
"pypinyin_g2p_phone",
"espeak_ng_arabic",
"espeak_ng_german",
"espeak_ng_french",
"espeak_ng_spanish",
"espeak_ng_russian",
"g2pk",
"g2pk_no_space",
],
default=None,
help="Specify g2p method if --token_type=phn",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
assert check_return_type(parser)
return parser
def add_task_arguments(cls, parser: argparse.ArgumentParser):
group = parser.add_argument_group(description="Task related")
# NOTE(kamo): add_arguments(..., required=True) can't be used
# to provide --print_config mode. Instead of it, do as
required = parser.get_default("required")
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(CepNet),
help="The keyword arguments for model class.",
)
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--input_size",
type=int_or_none,
default=None,
help="The number of input dimension of the feature",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=True,
help="Apply preprocessing to data or not",
)
parser.add_argument(
"--speech_volume_normalize",
type=float_or_none,
default=None,
help="Scale the maximum amplitude to the given value.",
)
parser.add_argument(
"--rir_scp",
type=str_or_none,
default=None,
help="The file path of rir scp file.",
)
parser.add_argument(
"--rir_apply_prob",
type=float,
default=1.0,
help="THe probability for applying RIR convolution.",
)
parser.add_argument(
"--noise_scp",
type=str_or_none,
default=None,
help="The file path of noise scp file.",
)
parser.add_argument(
"--noise_apply_prob",
type=float,
default=1.0,
help="The probability applying Noise adding.",
)
parser.add_argument(
"--noise_db_range",
type=str,
default="13_15",
help="The range of noise decibel level.",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def add_task_arguments(cls, parser: argparse.ArgumentParser):
group = parser.add_argument_group(description="Task related")
required = parser.get_default("required")
required += ["token_list"]
group.add_argument("--train_pseudo_shape_file",
type=str,
action="append",
default=[])
group.add_argument(
"--train_pseudo_data_path_and_name_and_type",
type=str2triple_str,
action="append",
default=[],
help=
"Give three words splitted by comma. It's used for the training data.",
)
group.add_argument(
"--token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token",
)
group.add_argument(
"--pis_ratio",
type=float,
default=1,
help="The multi-iteration ratio for SSL",
)
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--input_size",
type=int_or_none,
default=None,
help="The number of input dimension of the feature",
)
group.add_argument(
"--ctc_conf",
action=NestedDictAction,
default=get_default_kwargs(CTC),
help="The keyword arguments for CTC class.",
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetASRModel),
help="The keyword arguments for model class.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=True,
help="Apply preprocessing to data or not",
)
group.add_argument(
"--token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word", "phn"],
help="The text will be tokenized "
"in the specified level token",
)
group.add_argument(
"--bpemodel",
type=str_or_none,
default=None,
help="The model file of sentencepiece",
)
parser.add_argument(
"--non_linguistic_symbols",
type=str_or_none,
help="non_linguistic_symbols file path",
)
parser.add_argument(
"--cleaner",
type=str_or_none,
choices=[None, "tacotron", "jaconv", "vietnamese"],
default=None,
help="Apply text cleaning",
)
parser.add_argument(
"--g2p",
type=str_or_none,
choices=[None, "g2p_en", "pyopenjtalk", "pyopenjtalk_kana"],
default=None,
help="Specify g2p method if --token_type=phn",
)
parser.add_argument(
"--speech_volume_normalize",
type=float_or_none,
default=None,
help="Scale the maximum amplitude to the given value.",
)
parser.add_argument(
"--rir_scp",
type=str_or_none,
default=None,
help="The file path of rir scp file.",
)
parser.add_argument(
"--rir_apply_prob",
type=float,
default=1.0,
help="THe probability for applying RIR convolution.",
)
parser.add_argument(
"--noise_scp",
type=str_or_none,
default=None,
help="The file path of noise scp file.",
)
parser.add_argument(
"--noise_apply_prob",
type=float,
default=1.0,
help="The probability applying Noise adding.",
)
parser.add_argument(
"--noise_db_range",
type=str,
default="13_15",
help="The range of noise decibel level.",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def add_task_arguments(cls, parser: argparse.ArgumentParser):
group = parser.add_argument_group(description="Task related")
group.add_argument(
"--token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token",
)
group.add_argument(
"--src_token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token (for source language)",
)
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--input_size",
type=int_or_none,
default=None,
help="The number of input dimension of the feature",
)
group.add_argument(
"--ctc_conf",
action=NestedDictAction,
default=get_default_kwargs(CTC),
help="The keyword arguments for CTC class.",
)
group.add_argument(
"--enh_criterions",
action=NestedDictAction,
default=[
{
"name": "si_snr",
"conf": {},
"wrapper": "fixed_order",
"wrapper_conf": {},
},
],
help="The criterions binded with the loss wrappers.",
)
group.add_argument(
"--diar_num_spk",
type=int_or_none,
default=None,
help="The number of speakers (for each recording) for diar submodel class",
)
group.add_argument(
"--diar_input_size",
type=int_or_none,
default=None,
help="The number of input dimension of the feature",
)
group.add_argument(
"--enh_model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetEnhancementModel),
help="The keyword arguments for enh submodel class.",
)
group.add_argument(
"--asr_model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetASRModel),
help="The keyword arguments for asr submodel class.",
)
group.add_argument(
"--st_model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetEnhancementModel),
help="The keyword arguments for st submodel class.",
)
group.add_argument(
"--diar_model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetDiarizationModel),
help="The keyword arguments for diar submodel class.",
)
group.add_argument(
"--subtask_series",
type=str,
nargs="+",
default=("enh", "asr"),
choices=["enh", "asr", "st", "diar"],
help="The series of subtasks in the pipeline.",
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetEnhS2TModel),
help="The keyword arguments for model class.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=False,
help="Apply preprocessing to data or not",
)
group.add_argument(
"--token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word", "phn"],
help="The text will be tokenized " "in the specified level token",
)
group.add_argument(
"--bpemodel",
type=str_or_none,
default=None,
help="The model file of sentencepiece",
)
group.add_argument(
"--src_token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word", "phn"],
help="The source text will be tokenized " "in the specified level token",
)
group.add_argument(
"--src_bpemodel",
type=str_or_none,
default=None,
help="The model file of sentencepiece (for source language)",
)
group.add_argument(
"--non_linguistic_symbols",
type=str_or_none,
help="non_linguistic_symbols file path",
)
group.add_argument(
"--cleaner",
type=str_or_none,
choices=[None, "tacotron", "jaconv", "vietnamese"],
default=None,
help="Apply text cleaning",
)
group.add_argument(
"--g2p",
type=str_or_none,
choices=g2p_choices,
default=None,
help="Specify g2p method if --token_type=phn",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def add_task_arguments(cls, parser: argparse.ArgumentParser):
group = parser.add_argument_group(description="Task related")
# NOTE(kamo): add_arguments(..., required=True) can't be used
# to provide --print_config mode. Instead of it, do as
# required = parser.get_default("required")
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetEnhancementModel),
help="The keyword arguments for model class.",
)
group.add_argument(
"--criterions",
action=NestedDictAction,
default=[
{
"name": "si_snr",
"conf": {},
"wrapper": "fixed_order",
"wrapper_conf": {},
},
],
help="The criterions binded with the loss wrappers.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=False,
help="Whether to apply preprocessing to data or not",
)
group.add_argument(
"--speech_volume_normalize",
type=str_or_none,
default=None,
help="Scale the maximum amplitude to the given value or range. "
"e.g. --speech_volume_normalize 1.0 scales it to 1.0.\n"
"--speech_volume_normalize 0.5_1.0 scales it to a random number in "
"the range [0.5, 1.0)",
)
group.add_argument(
"--rir_scp",
type=str_or_none,
default=None,
help="The file path of rir scp file.",
)
group.add_argument(
"--rir_apply_prob",
type=float,
default=1.0,
help="THe probability for applying RIR convolution.",
)
group.add_argument(
"--noise_scp",
type=str_or_none,
default=None,
help="The file path of noise scp file.",
)
group.add_argument(
"--noise_apply_prob",
type=float,
default=1.0,
help="The probability applying Noise adding.",
)
group.add_argument(
"--noise_db_range",
type=str,
default="13_15",
help="The range of signal-to-noise ratio (SNR) level in decibel.",
)
group.add_argument(
"--short_noise_thres",
type=float,
default=0.5,
help=
"If len(noise) / len(speech) is smaller than this threshold during "
"dynamic mixing, a warning will be displayed.",
)
group.add_argument(
"--use_reverberant_ref",
type=str2bool,
default=False,
help="Whether to use reverberant speech references "
"instead of anechoic ones",
)
group.add_argument(
"--num_spk",
type=int,
default=1,
help="Number of speakers in the input signal.",
)
group.add_argument(
"--num_noise_type",
type=int,
default=1,
help="Number of noise types.",
)
group.add_argument(
"--sample_rate",
type=int,
default=8000,
help="Sampling rate of the data (in Hz).",
)
group.add_argument(
"--force_single_channel",
type=str2bool,
default=False,
help="Whether to force all data to be single-channel.",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def add_task_arguments(cls, parser: argparse.ArgumentParser):
group = parser.add_argument_group(description="Task related")
# NOTE(kamo): add_arguments(..., required=True) can't be used
# to provide --print_config mode. Instead of it, do as
required = parser.get_default("required")
required += ["token_list"]
group.add_argument(
"--token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token",
)
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--input_size",
type=int_or_none,
default=None,
help="The number of input dimension of the feature",
)
group.add_argument(
"--ctc_conf",
action=NestedDictAction,
default=get_default_kwargs(CTC),
help="The keyword arguments for CTC class.",
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetASRModel),
help="The keyword arguments for model class.",
)
group.add_argument(
"--use_curriculum",
type=str,
default=False,
help="Use curriculum learning algorithms.",
)
group.add_argument(
"--curriculum_algo",
type=str,
default="exp3s",
help="MAB algorithm to use for curriculum learning",
)
group.add_argument(
"--gain_type",
type=str,
default="PG",
help="Loss-based gain for Curriculum Learning. Prediction Gain (PG) or Self-Prediction Gain (SPG)",
)
group.add_argument("--refill_task",
type=bool,
default=1,
help="Refill task after the task is exhausted")
group.add_argument("--gen_log_dir",
type=str,
default='curriculum_log',
help="Directory to store curriculum generator logs")
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=True,
help="Apply preprocessing to data or not",
)
group.add_argument(
"--token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word", "phn"],
help="The text will be tokenized " "in the specified level token",
)
group.add_argument(
"--bpemodel",
type=str_or_none,
default=None,
help="The model file of sentencepiece",
)
parser.add_argument(
"--non_linguistic_symbols",
type=str_or_none,
help="non_linguistic_symbols file path",
)
parser.add_argument(
"--cleaner",
type=str_or_none,
choices=[None, "tacotron", "jaconv", "vietnamese"],
default=None,
help="Apply text cleaning",
)
parser.add_argument(
"--g2p",
type=str_or_none,
choices=[None, "g2p_en", "pyopenjtalk", "pyopenjtalk_kana"],
default=None,
help="Specify g2p method if --token_type=phn",
)
parser.add_argument(
"--speech_volume_normalize",
type=float_or_none,
default=None,
help="Scale the maximum amplitude to the given value.",
)
parser.add_argument(
"--rir_scp",
type=str_or_none,
default=None,
help="The file path of rir scp file.",
)
parser.add_argument(
"--rir_apply_prob",
type=float,
default=1.0,
help="THe probability for applying RIR convolution.",
)
parser.add_argument(
"--noise_scp",
type=str_or_none,
default=None,
help="The file path of noise scp file.",
)
parser.add_argument(
"--noise_apply_prob",
type=float,
default=1.0,
help="The probability applying Noise adding.",
)
parser.add_argument(
"--noise_db_range",
type=str,
default="13_15",
help="The range of noise decibel level.",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
